PSPNet

This is an unofficial implementation of Pyramid Scene Parsing Network (PSPNet) in Chainer.

Training

Requirement

• Python 3.4.4+
  • Chainer 3.0.0b1+
  • ChainerMN master
  • CuPy 2.0.0b1+
  • ChainerCV 0.6.0+
  • NumPy 1.12.0+
  • tqdm 4.11.0+

pip install chainer --pre
pip install cupy --pre
pip install git+git://github.com/chainer/chainermn
pip install git+git://github.com/chainer/chainercv
pip install tqdm

---

Inference using converted weights

Requirement

• Python 3.4.4+
  • Chainer 3.0.0b1+
  • ChainerCV 0.6.0+
  • Matplotlib 2.0.0+
  • CuPy 2.0.0b1+
  • tqdm 4.11.0+

1. Run demo.py

Cityscapes

$ python demo.py -g 0 -m cityscapes -f aachen_000000_000019_leftImg8bit.png

Pascal VOC2012

$ python demo.py -g 0 -m voc2012 -f 2008_000005.jpg

ADE20K

$ python demo.py -g 0 -m ade20k -f ADE_val_00000001.jpg

FAQ

If you get RuntimeError: Invalid DISPLAY variable, how about specifying the matplotlib's backend by an environment variable?

$ MPLBACKEND=Agg python demo.py -g 0 -m cityscapes -f aachen_000000_000019_leftImg8bit.png

---

Convert weights by yourself

Caffe is NOT needed to convert .caffemodel to Chainer model. Use caffe_pb2.py.

Requirement

• Python 3.4.4+
  • protobuf 3.2.0+
  • Chainer 3.0.0b1+
  • NumPy 1.12.0+

1. Download the original weights

Please download the weights below from the author's repository:

• pspnet50_ADE20K.caffemodel: GoogleDrive
• pspnet101_VOC2012.caffemodel: GoogleDrive
• pspnet101_cityscapes.caffemodel: GoogleDrive

and then put them into weights directory.

2. Convert weights

$ python convert.py

---

Reference

• The original implementation by authors is: hszhao/PSPNet
• The original paper is:
  • Hengshuang Zhao, Jianping Shi, Xiaojuan Qi, Xiaogang Wang, Jiaya Jia, "Pyramid Scene Parsing Network", Proceedings of IEEE Conference on Computer Vision and Pattern Recognition (CVPR), 2017